A new strategy in cancer prognosis is to base the decision on integrated information from different sources, including the traditional clinical and demographic information of patients, such as age, grade, and tumor size, etc, and the recently emerged genetic information like expression of gene or protein markers. Implementation of such a strategy requires efficient quantitative models that integrate the clinical measurements and genetic measurements together for prognosis. The long-range goal of this application is to improve risk predication, treatment selection, and subtype classification in cancer prevention, diagnosis, and prognosis. The short-term objective is to improve prediction of treatment response for cancer patients by developing innovative statistical models that integrate three different types of data, including two subtypes of informatics data, namely protein pathway data and high-throughput protein expression data, and a third type, which is the standard clinical and demographic data. We will accomplish the objective of this application by pursuing the following five specific aims: 1) Develop Bayesian parametric models that integrate a known genetic pathway with high-throughput protein expression measurements. 2) Develop Bayesian nonparametric model that integrate multiple genetic pathways with protein expression measurements. 3) Develop Bayesian classification procedures based on the Bayesian models proposed in previous two aims. 4) Integrate clinical and demographic measurements into the Bayesian models and apply the Bayesian classification procedures using a comprehensive data set that contains protein expression measurements and clinical measurements for more than 500 patients with leukemia. 5) Validate statistical findings by performing biological experiments, which will be done by our collaborating biologists. The proposed research is expected to provide quantitative prognostic tools for oncologists based on integrated information. The impact of the proposed research will be significant because models developed in this application can be applied to various cancer types and thus potentially improve the prognosis for patients with different types of cancer. PUBLIC HEALTH RELEVANCE: Integrating the protein expression data, the protein pathway data, and the clinical data is expected to significantly improve medical decision making such as treatment selection. The improved decisions are expected to improve the overall patient care. For example, by accurately predicting that certain treatment will not be effective for a cancer patient, this patient will no longer waste time trying out the treatment and will have a better chance finding some other more effective therapies.